Cooperative and noncooperative binding of protein ligands to nucleic

James G. McAfee, Syrus R. Soltaninassab, Mark E. Lindsay, and Wallace M. LeStourgeon. Biochemistry 1996 35 (4), 1212-1222. Abstract | Full Text HTML |...
0 downloads 0 Views 2MB Size
1226

Biochemistry 1986, 25, 1226-1240

Neya, S., & Morishima, I. (1981) J . Biol. Chem. 256, 793-798. Neya, S., Hada, S., & Funasaki, N. (1983) Biochemistry 22, 3686-3691. Neya, S., Hada, S., & Funasaki, N. (1985a) Biochim. Biophys. Acta 828, 241-246. Neya, S., Hada, S., Funasaki, N., Umemura, J., & Takenaka, T. (1985b) Biochim. Biophys. Acta 827, 157-163. Perutz, M. F., Heidner, E. J., Ladner, J. E., Bettlestone, J. G., Ho, C., & Slade, E. F. (1974) Biochemistry 13, 21 87-2200. Perutz, M. F., Sanders, J. K. M., Chenery, D. H., Noble, R. W., Pennelly, R. R., Fung, L. W.-M., Ho, C., Giannini, F., Porschke, D., & Winkler, H. (1978) Biochemistry 17, 3640-36 52. Philo, J. S., & Dreyer, U. (1985) Biochemistry 24,2985-2992. Savicki, J. P., Lang, G., & Ikeda-Saito, M. (1984) Proc. Natl.

Acad. Sci. U.S.A. 81, 5417-5419. Scholler, D. M., & Hoffman, B. M. (1979) J . Am. Chem. SOC. 101. 1655-1662. Scholler, D. M., Wang, N.-Y. R., & Hoffman, B. M. (1 978) Methods Enzymol. 52, 487-493. Verloop, A., Hoogenstraaten, W., & Tipker, J. (1976) in Drug Design (Ariens, E. J., Ed.) Vol. 7, pp 165-207, Academic Press, New York. Winter, M. R. C., Johnson, C. E., Lang, G., & Williams, R. J. P. (1972) Biochim. Biophys. Acta 263, 515-534. Wuthrich, K. (1970) in Structure and Bonding (Hemmerich, R., Jorgensen, C. K., Neilands, J. B., Nyholm, S. R. S., Reinen, D., Williams, R. J. P., Eds.) Vol. 8, pp 53-121, Springer-Verlag, West Berlin. Yonetani, T., Iizuka, T., Asakura, T., Otsuka, J., & Kotani, M. (1972) J . Biol. Chem. 247, 863-868.

Cooperative and Noncooperative Binding of Protein Ligands to Nucleic Acid Lattices: Experimental Approaches to the Determination of Thermodynamic Parameters+ Stephen C. Kowalczykowski,**gLeland S . Paul,**’, Nils Lonberg,t,i John W. Newport,$,# James A. McSwiggen,txo and Peter H . von Hippel*,* Institute of Molecular Biology and Departments of Chemistry and Biology, University of Oregon, Eugene, Oregon 97403, and Department of Molecular Biology, Northwestern University Medical School, Chicago, Illinois 6061 1 Received August 2, 1985

ABSTRACT: Many biologically important proteins bind nonspecifically, and often cooperatively, to singleor double-stranded nucleic acid lattices in discharging their physiological functions. This binding can generally be described in thermodynamic terms by three parameters: n, the binding site size; K , the intrinsic binding constant; w , the binding cooperativity parameter. The experimental determination of these parameters often appears to be straightforward but can be fraught with conceptual and methodological difficulties that may not be readily apparent. In this paper we describe and analyze a number of approaches that can be used to measure these protein-nucleic acid interaction parameters and illustrate these methods with experiments on the binding of T4-coded gene 32 (single-stranded D N A binding) protein to various nucleic acid lattices. We consider the following procedures: (i) the titration of a fixed amount of lattice (nucleic acid) with added ligand (protein); (ii) the titration of a fixed amount of ligand with added lattice; (iii) the determination of ligand binding affinities at very low levels of lattice saturation; (iv) the analysis of ligand cluster size distribution on the lattice; (v) the analysis of ligand binding to lattices of finite length. The applicability and limitations of each approach are considered and discussed, and potential pitfalls are explicitly pointed out.

T e nonspecific binding of proteins to single- or doublestranded nucleic acid lattices is a central feature of many ‘These studies were supported by USPHS Research Grants GM15792 and GM-29158 (to P.H.v.H.), as well as (in part) by Research Grant AI-I8987 (to S.C.K.). L.S.P., J.W.N., and J.A.M. were predoctoral trainees on USPHS Institutional National Research Service Award GM-07759 during parts of these studies. Some of this work was also submitted by L.S.P. to the Graduate School of the University of Oregon in partial fulfillment of the requirements for the Ph.D. degree in chemistry. *University of Oregon. 5 Northwestern University Medical School. Present address: Abbott Laboratories, North Chicago, IL 60064. Present address: Sloan-Kettering Institute, New York, N Y 10021. # Present address: Department of Biology, University of California at San Diego, La Jolla, CA 92093. O Present address: Department of Molecular, Cellular, and Developmental Biology, University of Colorado, Boulder, CO 80309.



0006-2960/86/0425- 1226$01.50/0

functional and regulatory biological processes. Theoretically, the problem can be viewed as the binding of large ligands to a one-dimensional lattice, where each ligand interacts with more than one lattice unit (nucleotide residue or base pair), and thus also covers more than one potential ligand binding site (Le., ligand binding is of the “overlap” type). Overlap binding of ligands to a one-dimensionallattice complicates the analysis of titration curves because the binding sites on the lattice are not titrated independently. Because of overlap, the number of free lattice binding sites occluded per binding event decreases with increasing saturation of the lattice. As a consequence, overlap binding is effectively “negatively cooperative”, and it becomes progressively more unfavorable to bind additional ligands as lattice saturation is approached. In addition, protein binding to nucleic acid lattices may also be (and generally is) positively cooperative,in that the binding

0 1986 American Chemical Society

VOL. 2 5 , NO. 6 , 1986

BINDING PROTEINS TO NUCLEIC ACID LATTICES

of an additional protein adjacent to one that is already bound is favored over binding to an isolated site. This antagonizes the overlap effect, promotes the clustering of bound ligands, and facilitates lattice saturation. [For a complete theoretical discussion of overlap and cooperative binding to a linear lattice, see McGhee & von Hippel (1974).] The effects of nonspecific protein-nucleic acid interactions on biological regulatory mechanisms are widespread and can be illustrated by the following examples. The processes of D N A replication, recombination, and repair require the binding of single-stranded D N A binding proteins to transiently single-stranded sequences formed as intermediates in these processes [for a review, see Kowalczykowski et al. (1981a)l; binding is nonspecific and often cooperative to force saturation of the lattice. Repressors, R N A polymerase, and other genome regulatory proteins use nonspecific binding to random sequences on the chromosome as a n intermediate step in the location of specific D N A regulatory targets such as operators and promoters (Berg et al., 1982) and to provide modulation of the net affinity for the specific binding site through competitive binding (von Hippel et al., 1974; Lin & Riggs, 1975). Nonspecific protein-nucleic acid interactions have also been utilized in controlling the translation of messenger R N A through the cooperative binding of the regulator protein to the ribosome binding site on the R N A [see von Hippel et al. (1982)l. Analysis of the above systems (as well as other functional and regulatory systems involved in gene expression) has shown that quantitative measures of nonspecific proteinnucleic acid binding parameters are required to interpret (and to discriminate) competing molecular biological models of such processes, since such models can often be distinguished only in physical chemical terms. A variety of mathematical approaches have been taken to solving the problem of nonspecific binding of large ligands to one-dimensional lattices. Latt and Sober (1967) used both a combinatorial method and the method of sequence-generating functions to investigate lattice binding of noninteracting ligands. Crothers (1968) and Schwarz (1970) used a matrix method to describe both the noninteractive and the interactive (cooperative) binding of ligands to nucleic acid lattices. Zasadetlev et al. (197 1) developed a combinatorial procedure for the same purposes, while Schellman (1974) has applied the method of sequence-generating functions to the solution of this problem. McGhee and von Hippel (1 974) used a conditional probability approach and were able to obtain solutions in closed analytic form for both cooperative and noncooperative binding. Despite this multiplicity of theoretical approaches, the problems associated with applying these theories to real systems have often led to difficulty and confusion. In our laboratories, over a period of years, we have been developing a variety of experimental approaches to these protein-nucleic acid interaction systems and have been trying to sort out the interpretive complications associated with each. Some of these approaches are summarized here. For convenience, we have formulated our discussion in the context of the McGhee-von Hippel version of the binding theory, which has the advantage of easy computational manipulation. However, this presentation could easily be recast in terms of any of the other theoretical methods listed above [e.g., see Schwarz & Watanabe (1983) and Watanabe & Schwarz (1983)l. The simple model used in all these theories, as well as in this paper, describes the binding of proteins to nucleic acids in terms of three thermodynamic parameters: n, the binding site size of the protein (in units of nucleotide residues or base pairs); K , the intrinsic binding constant (in M-’);’ w , the co-

1

1227

4

Kw

I +&+$- - I

Isolated Binding

Contiguous Binding

I

FIGURE1: Definitions of the thermodynamic parameters that describe the interaction of a protein with a nucleic acid lattice. Each arrowhead represents a lattice site (Le., a nucleotide residue), and the illustrated protein ligand covers three such sites (n = 3). K (M-I) is the intrinsic association constant for protein binding to the lattice at an isolated site, and w (dimensionless) represents the cooperativity of binding. w is defined as the equilibrium constant for moving a protein from a given isolated site to a given contiguous binding site. Thus, K u is the net binding constant per contiguously bound protein molecule. If contiguous binding is favored, w > 1, if the binding is noncooperative, w = 1, and if the binding is negatively cooperative, w < 1.

operativity parameter (unitless) that specifies the relative affinity of an incoming ligand for a contiguous, as opposed to a n isolated, binding site. These parameters are illustrated in Figure 1. Obviously a description in terms of single values of n, K , and w provides an oversimplified model of many real biological systems. Even the paradigm system (T4gene 32 protein binding to single-stranded DNA) fails to fit it exactly. Possible complications include heterogeneous binding to compositionally different nucleic acid sequences (more than one value of K ) , different modes of binding of the protein to the nucleic acid lattice (more than one value of n), and cooperative interactions between ligands beyond the nearest-neighbor level (more than one value of w). Furthermore, most of the theoretical models (including McGhee-von Hippel) were developed to work with long (effectively infinite) lattices, while many real biological systems involve binding to short nucleic acid lattices. Theoretical ways of handling the finite lattice problem have been developed by Epstein (1978) and by J. A. Schellman (unpublished results); some experimental consequences of finite lattice binding are discussed in this paper. The descriptive limitations of the model we use should be clearly borne in mind. However, the “open-endedness” of assigning more detailed and specific interaction parameters to models of real biological systems, plus the errors involved in real measurements, generally limits us to the use of the simple model and the three thermodynamic parameters defined above. The determination and manipulation of these parameters is described in what follows. Knowing these parameters for a given system is generally useful in itself, and obviously they can be further refined or subdivided as additional knowledge of the system a t issue is gathered by other techniques. The units of K , like those of n, can be expressed either as nucleotide residues or as base pairs. The former represents the obvious choice for the formation of the single-stranded nucleic acid-protein complexes that are primarily considered as examples in this paper. Either set of units is appropriate for the treatment of double-stranded nucleic acid-protein complexes, but the choice must be explicitly made and stated since the magnitude of the listed value of K (and of n) will depend on it. For additional discussion of this point, in connection with E . coli lac repressor-DNA complexes, see Butler et al. (1977).

1228

BIOCHEMISTRY

KOWALCZYKOWSKI ET AL.

MATERIALS AND METHODS The preparation and manipulation of the nucleic acids and proteins used in this work, as well as the details of experimental procedures, specific buffer systems, etc., have been described elsewhere (Kowalczykowski et al., 1981b;Newport et al., 1981; Lonberg et al., 198 1). The fluorescence titrations described in this paper were performed either with a Schoeffel spectrofluorometer, as described by Kowalczykowski et al. (1981b), or with a SLM-8000 spectrofluorometer interfaced to a Hewlett-Packard HP-87 computer (J. A. McSwiggen et al., unpublished results). Fractionated samples of poly(riboadenylic acid) [poly(rA)] were obtained from Miles Biochemical Co. EXPERIMENTAL APPROACHES

0

Equations Because the equations derived by McGhee and von Hippel (1 974) can be written in closed form, they provide a useful basis for the quantitative analysis of the binding of proteins to nucleic acids. For noncooperative systems, we write V

L

0

I

2

3

4

Free Ligand Concentration ( M x IO7)

n- 1

=

K(l

-

nu)

(1)

and for cooperative systems, we write

1 = ~ ( -1 nu) x L

[

(2w - 1)(1

- nv)

2 ( w - 1)(1

+v -R

- nu)

I*'[

1-(n+l)v+R 2(1 - n u )

I

(2)

+

+

where R = [[l - (n l)v]* 4 4 1 - n ~ ) ] ' v/ is ~ ,the binding density of the ligand on the lattice (in moles of ligand per mole of nucleotide residue or base pair), and L is thefree ligand concentration (in moles per liter). The derivation of eq 1 and 2 assumes that ligand binding occurs on an infinite lattice. W e use these equations throughout because of their ease of interpretation and evaluation. In a later section, we examine the limitations for real systems that result from the infinite lattice assumption. Titration of Lattice with Ligand The titration of a fixed amount of lattice with added ligand is probably the most direct and commonly used method of determining the thermodynamic parameters of a ligand-lattice (protein-nucleic acid) interaction. Typically, some latticedependent spectroscopic signal, reflecting a change in a lattice property that varies linearly with ligand binding, is monitored as a function of total ligand concentration. Less commonly, a ligand-based signal that is perturbed by the binding interaction is used, and the change in this signal is used to monitor the progressive saturation of the lattice. For our measurements of gene 32 protein binding, changes in the UV absorbance, circular dichroism, or fluorescence of the nucleic acid or of an appropriate nucleic acid derivative have been monitored as protein is added; alternatively, the deviation from a linear increase of the intrinsic fluorescence of the protein (due to quenching of the protein fluorescence in the bound state) as it is added to a lattice-containing solution may be followed [see Alma et al. (1 983) for an example of this approach]. The result of a titration of lattice with ligand generally resembles the family of curves shown in Figure 2. In this figure we present model titration curves, which have been generated with eq 1 and 2, for noncooperatively and cooperatively binding ligands. The total apparent affinity (i.e., the

0

I

2

3

4

Total Ligond Concentration ( M x IO7) FIGURE 2: (a) Fractional lattice saturation vs.free ligand concentration for various values of K and w ; the product of K and w has been held constant at 1 X lo7 M-'. Values of w are indicated in the figure. The site size (n)is 10, and the concentration of lattice is 0. (b) Plot identical M, and the x with (a), but the concentration of lattice units is axis becomes total ligand concentration.

product of K and w ) has been held constant in these calculations, while the individual values of K and w are varied reciprocally. T$e most prominent feature of these plots is the progressive debelopment of the classical sigmoid titration curve (indicative of cooperative ligand binding) as w is increased. In addition, note that all of the curves with values of (L! significantly larger than n pass through a common value that falls a t the midpoint of each titration curve. As was pointed out by McGhee and von Hippel (1974),the value of the free protein concentration at the midpoint of such binding isotherms approaches 1/(Kw) (for curves plotted in terms of free ligand concentration) for values of w that are significantly larger (- 10-fold) than n; the usefulness of this relationship will become apparent when the analysis of cooperatively binding systems is discussed. The curves of Figure 2 have been presented in two equivalent ways in order to illustrate the difference between plotting fractional saturation as a function of free ligand and of total ligand concentration. Although the overall shapes of the plots are similar, the slopes of the steeply rising portions of the cooperative curves differ, due solely to the fact that in Figure 2b the total ligand concentration includes the ligand that is bound to the lattice. The amount of bound ligand is simply equal to uN, where N is the total concentration of lattice units (in nucleotide residues) present and v (the binding density)

VOL. 2 5 , NO. 6 , 1986

BINDING PROTEINS T O NUCLEIC ACID LATTICES

-0

08

16

2'4

j2

1229

(1981a). In the next sections, we discuss approaches for the quantitative analysis of titrations of lattice with added ligand for systems in which binding is nonstoichiometric and either noncooperative or cooperative. The Noncooperative Case. When w has a value close to 1, no sigmoidicity of the binding isotherm is observed (see Figures 2 and 3). It is then a simple matter to model the titration curve if a value of n has been obtained from a stoichiometric binding experiment as described above. With this value of n, theoretical curves can be generated for various trial values of the binding constant ( K ) with eq 1 and compared with the experimental results. Alternatively, a nonlinear fitting procedure can be used (see Appendix). It is possible also to determine the value of K from the midpoint of the titration curve (Le., at 8 = By substituting the value of n, the value of the free ligand concentration at the midpoint of the titration ( L I l 2 )and , the value of v at the midpoint [ v ~ ,which ~ , equals 1/(2n)], into eq 1, we obtain

Free Ligand Concentration ( M x IO7]

FIGURE 3: Fractional lattice saturation vs. free ligand concentration

for a noncooperatively binding ligand, at various values of K,with n = 10 and w = 1.

is equal to the fractional saturation of the lattice (8) divided by the ligand binding site size (n). Thus, the two sets of curves in Figure 2 are related through conservation of mass by .C, = L + vN = L 8 N / n (3) where LT is the total ligand concentration and L is the free ligand concentration. Since experimental data are obtained as a function of total ligand concentration and the theoretical curves generated with eq 1 and 2 are in terms of thefree ligand concentration, it is necessary to convert one form to the other. However, in order to use eq 3 for this purpose, the binding site size (n) must be obtained, since 8, rather than v, is generally the preferred dependent variable.2 The site size (n) for a protein can often be obtained by conducting a titration under stoichiometric binding conditions. In such a titration, an abrupt "break" is seen at the equivalence point, and the molar ratio of lattice residues (e.g., nucleotide residues or base pairs) to ligand concentration at the break is equal to the site size [for real examples, see Figure 3 in Kowalczykowski et al. (198 1b) or Figure 1 in Newport et al. (1 98 l)]. The problem of defining stoichiometric binding conditions will be addressed specifically below. However, such conditions can generally be defined experimentally as those under which there is no change in the stoichiometry at the equivalence point when the lattice concentration is varied over an approximately 10-fold range. Because the binding affinity of proteins for nucleic acids is often dependent on salt concentration [see Record et al. (1978, 1981) for reviews], one of the easiest experimental ways to increase the binding affinity of the protein for the nucleic acid is to lower the salt concentration. Measured site sizes, and the conditions required for tight binding for a number of single-stranded DNA binding proteins, have been summarized in Kowalczykowski et al.

+

* In principle, it is possible to determine n, K , and w simultaneously from a threeparameter fit of the experimental data to eq 1 and 2. While the titration curve itself should contain sufficient information to define these parameters uniquely, we have found that in practice an unambiguous (uncorrelated) fit generally cannot be obtained, due to the experimental uncertainty of the data. Thus, we generally determine the site size (n) first and then fit the experimental data to trial values of K and w . In this way, only a two-parameter fit is required, which (as will be demonstrated below) can often be reduced to a one-parameter fit under special circumstances.

nLIj2(1+ I/n)"-' (4) This equation can be used to determine exact values for K i f n is known. If n is not known exactly, an estimate of the binding constant (Kapprox) can be obtained by ignoring ligand overlap, thus reducing eq 4 to

Kapprox = 1/ W I / 2 )

(5) For example, for n = 10, eq 5 underestimates the true value of K approximately 2.3-fold, and for n = 5, it underestimates K 2-fold. Nevertheless, eq 5 is often helpful in estimating trial values of K for use in curve-fitting methods (see Appendix). Despite the relative simplicity of determining K from noncooperative titration data once n is known, two experimental problems often make the analysis more difficult. The first is encountered when conditions cannot be found under which ligand binding is stoichiometric, thus precluding an independent determination of n. The second is related to the first, in that if binding is weak, or if n is large, saturation of the lattice cannot be achieved due to overlap problems [see McGhee & von Hippel (1974)] and consequently the end point of the titration cannot be accurately determined. In Figure 3, we have plotted a number of curves with correspondingly weaker binding affinities from left to right; such data might be encountered in protein-nucleic acid titrations as the salt concentration is increased [e.g., see Lonberg et al. (1981); Figure 61. In the absence of an independent value of n, the simple one-parameter determination of K from a noncooperative titration curve becomes a three-parameter problem, Le., to determine n, K . and the true end point. Other than a direct three-parameter fit, there is no easy solution to this situation. We have had some limited success by using a "bootstrap" approach whereby the experimental end point is assumed to represent some value of 8 less than 1, and the curve is then fit to values of n and K . These first approximations to n and K are then used to calculate the actual value of 0 at the end point, and this value is used rather than the initially assumed 0 value to redetermine a "second-order" fit of n and K , and so forth. This procedure is used iteratively until n, K , and 0 no longer change (see Appendix). The Cooperative Case. When binding is highly cooperative (Le,, titration curves are noticeably sigmoidal, which indicates

-

KOWALCZYKOWSKI ET AL,

1230 B I O C H E M 1ST R Y

-

1-

l , _

0

oe

I6

32

24

0

40

Fraa Lipand Concanriotion I M I IO')

Fractional lattice saturation va. free ligand concentration for a oooperatively binding ligand with w = Id,n = 10, and K varying as indicated. FIGURE 4:

that w 2 Ion), the determination of n, K, and w is relatively straightforward. First, n is determined from a stoichiometric titration curve as described previously. In the absence of such data, a good estimate of n can be obtained from the region of the binding curve that is steeply increasing. This portion of the curve basically represents the amount of protein nceded to titrate the lattice. As can be mn in Figure 2b, that binding (for the most cooperative case) begins at 1 X lo-' M ligand and reaches saturation at 2 X IO-' M. Since the total lattice concentration in these model plots was 1 X lOa M,with a site size (n) of 10, this method of estimating bound ligand is clearly quite good for highly cooperative titrations. It becomes less accurate with smaller values of w, but as can be seen in Figure 2b, it is still -50% accurate with values of w as low as IO3. As binding conditions become more nonstoichiometric (e& as the salt concentration is raised), three different classes of changes may occur: (i) K may decrease with w remaining constant; (ii) w may decrease with K remaining constant; (iii) both K and w may decrease. Class i is the most common for systems so far studied and has been observed with T4 gene 32 protein (Kowalczykowski et at., 1981b; Newport et al., 1981). fd gene 5 protein (Alma et al., 1983;Fomchke & Rauh, 1983), and Escherichia coli recA protein (Menetski & Kowalczykowski, 1985); qualitatively, a family of curves such as that of Figure 4 is observed. Notice that the shape of each curve is about the same, but that they are shifted to the right as K decreases. For class ii, behavior such as that of Figure 5 is observed. Here, each curve displays a "lag" region, but the slop of the steep portion of each decreases as w is decreased. The observed behavior for class iii will be a mixtur: of classes i and ii; Le., the curves shift to the right and the slopes decrease. The actual extent of each effect will depend on which parameter is changing more rapidly with the independent variable that is being manipulated. In order to extract values of K and w from curves such as these of Figures 4 and 5, we take advantage of the fact (see above) that when w is much larger than n, the value of the product of K and w is defined by the value of I/t(the reciprocal of the free ligand titration) at the midpoint of a nonstoichiometric titration (sac Figure 2a). Note that if the binding isotherm is plotted in terms of rord (rather thanfree) ligand concentration, then KO is related to the total ligand concentration at the midpoint, lq,,12,by

OB

24

16

ham Lqond Concmlratm LM

K

101 '

32

40

Same as Figure 4 except that K is fixed at 104 M" and varies as indicated.

FIGURE 5:

w

-

1 /(K4

-

LT,1/2

-N/(W

(6)

This equation is obtained by rearranging cq 3 and setting 8

0

!-0

-7 0 10

r---- -

0 20

1

----

OX,

c 3

[Gone 32 ProteinIpM

FIGURE 6: Titration of poly(riboethenoadcny1ic acid) (lob M residues) with gene 32 protein in buffer B [see Kowalczykowski et al. (1981b)l containing 0.45 M NaCl at 25 OC. The solid lines are theoretical curves for various values of w (calculated with Kw held constant at 7.0 X lo6 M-'and n = 7).

= the second term on the right in eq 6 is simply the concentration of ligand that is bound at the midpoint 7f the titration. Once the product KO is known, separate values of K and w can be determind 5y generating theoretical titration curves, with eq 2, in which K and w are varied individually. An example of this procedure is shown in Figure 6,with data on the binding of T4 gene 32 protein to poly[ribo(etheno)adenylic acid]. The solid lines in Figure 6 represent theoretical curves that have been obtained by incrementing values of Y in eq 2 and solving for L in terms of given values of n, K, and w. Note that the data points (unlike the theoretical curves) are not perfectly symmetrical about the midpoint of the plot. Thus, one model curve cannot be used to fit both the lower and upper portions of the experimental data. This follows because the polynucleotide lattices are not really of infinite length (as asaumed in the binding theory), so that "end effects" and lattice-length heterogeneity effects become important when the lattice is more than half-saturated. More will be said of finite lattiocs below, but h e n we emphasize that if the data are nor symmetrical about the midpoint of the titration, the lower "halt" of the titration should be emphasized in fitting procedures to minimize such problems. From a comparison of the data and model curves in Figure 6, one can sec that values of w < 500 produce binding isotherms that are too shallow to fit the data, while bindiiq

VOL. 25, NO. 6,

B I N D I N G PROTEINS TO N U C L E I C A C I D LATTICES

isotherms corresponding to values of w greater than 1 X lo4 are too steep. Thus, a range of fits can be readily established, with the “best” fit obtained for this system at w = 1.5 X lo3. We note that this procedure loses resolution when values of w become very large; under these conditions, only a minimum estimate of w is obtained. The raw titration data can be replotted in terms of free ligand concentration by simply subtracting the amount of ligand bound at each point from the total ligand concentration, with eq 3. This has the advantage of making the slope of the steeply rising portion of the observed titration curve independent of polynucleotide concentration (see Figure 4; Kowalczykowski et al., 1981b), and therefore, data sets obtained at different concentrations of lattice can be presented on the same plot. Either method of graphical presentation of the data is suitable for analysis. Instances where the cooperativity parameter is only on the order of 1-10-fold greater than n are the most difficult to fit accurately, because eq 6 does not apply. Here, at least a two-parameter fit is required, even if the site size is independently known. Although eq 6 may not provide an accurate approximation of the product Kw, as seen in Figure 2a for the curves representing values of w of 1 and 10, it may still be useful in obtaining a starting point for further numerical approximation. If the data are quite good, a two-parameter fit will be reasonably accurate, although difficulty in assigning unique values of K and w may be encountered because the two parameters are correlated. An independent measure of the binding constant can be obtained by performing binding measurements at low ligand saturation densities (see below).

Titration of Ligand with Lattice Another relatively common way of performing a titration-often made necessary by solubility or stability limits of the protein in the absence of nucleic acid-is to add increasing amounts of lattice to a fixed concentration of ligand. Again, spectroscopic changes in ligand properties can be used to measure interaction, but in such titrations the data are usually plotted in terms of fractional saturation of ligand (instead of lattice as in the previous section), since here a fixed amount of ligand is being titrated. As an example, the binding of polynucleotide by gene 32 protein results in quenching of the intrinsic fluorescence of the protein, and thus, this signal can be used to monitor the extent of ligand binding. Although this is an experimentally straightforward procedure, the analysis of the results obtained from such titrations for cooperatively binding systems is not straightforward and can easily result in significant misinterpretations. The reason for this is qualitatively that while the free ligand concentration continuously increases in a ligandto-lattice titration, in a lattice-to-ligand titration one begins with a great excess of ligand and ends with a great excess of lattice. This means, for highly cooperative systems, that as lattice is added and free ligand concentration in the titration cell is depleted a point will be reached at which the concentration of ligand that remains free is only l/(Kw). As more lattice is added beyond this point, additional ligand will only bind noncontiguously (Le., with a net affinity of K rather than of KO; Figure 1). As a result, an apparent plateau in the titration will be reached at a level that will depend on the total ligand concentration used in the titration. This plateau will not correspond to saturation. To illustrate this effect quantitatively, in Figure 7 we present the results of some model calculations that are based on eq 1 and 2 and use concentration values that might be encountered in a typical fluorescence titration. In order to present

1986

1231

n-IO rLlgond1 =

z

5

IO-~M

L--J ~~

m

0

04

00

12

-

~

16

20

Total Lattice Concentration ( x IO5)

FIGURE7 : Fractional ligand saturation vs. total lattice concentration for various values of K and o; the product of K and o was held constant at lo7 M-’. The values of w are indicated in the figure. The site size is 10, and the concentration of total ligand is 2.5 X lo-’ M.

curves with a wide range of u values in this figure, the net affinity (KO) is held constant at lo7 M-l, but the individual values of K and w are varied as in Figure 2. It is apparent that although the same values for n, K, and w were used as in the ligand-to-lattice titration of Figure 2, the curves in Figure 7 bear no resemblance (except for w = 1) to those of Figure 2. Note that as the value of w is increased (and K correspondingly decreased) two effects on the titration curves are observed: (i) the initial regions of the curve become sharper, as expected; (ii) the curves seem to approach a plateau value that corresponds to less than 100% ligand saturation. Although the more cooperative curves appear to be saturating at a plateau value representing less than 100% binding, this “plateau” is, in fact, gradually increasing and eventually approaches 100% saturation at very high concentrations of lattice (data not shown). However, this “real saturation” region may be experimentally inaccessible and thus may never be reached. As pointed out above, the basis of this apparent “plateau effect” in cooperative titrations is easily ubderstood when one realizes that when lattice is added to ligand, titrations such as those in Figure 2 are being performed “in reverse”. That is, the binding density of the lattice ( u ) is being decreased as lattice is added, so that in Figure 2 one is proceeding along the binding isotherm from right to left. When the free ligand concentration is depleted to the point where there is little ligand binding in Figure 2a [leftmost side of the isotherms where the concentration of free protein is less than l / ( K w ) ] ,an apparent end point is reached. This reasoning suggests that the percentage bound at the “apparent” plateau can be increased simply by increasing the input (total) ligand concentration, so that when the apparent end point at a ligand concentration of 1 / ( K w ) is reached, the bound ligand represents a smaller percentage of the total ligand concentration. Figure 8 shows the effect of varying the total concentration of ligand on the level of the apparent “plateau”. All the parameters of Figure 8 are identical with those of Figure 7, except that the initial protein concentration is varied. Comparison of these figures shows that increasing the input concentration of ligand also increases the apparent plateau level for these cooperatively binding systems. The curves begin to approach the true value of saturation (100%) only when the input ligand concentration exceeds the value of 1/ ( K w ) by approximately 10-fold. Note in Figure 8 that these effects are not seen for a noncooperatively binding ligand or for one that binds moderately cooperatively (w = n); this can be understood by observing that when w = 1-10, there is no distinct

1232

B I oc H E M I ST R Y 10

KOWALCZYKOWSKI ET AL.

K~ = IO’M-’

----- - _ _ _ - -r ; - ,A

[Ligand]= I O x .* - A

I

5,

0 x IO-M ’/

/

i

~

/’

2.5 IO-’U

-

04

0

12

00

7

-

16

1

20

Total Lattice Concentrotion ( x IO5) [Ligand]

0

:5

x Id’M

05 IO 15 Total L o t t i c e Concentration ( x IO5)

t

20

FIGURE 9: Effect of varying ligand concentration in a titration of ligand by lattice for a highly cooperative system; n = 10, K = 1 M-I, and w = 10’. The concentration of ligand is indicated in the figure. The breakpoints of the titration curve lie on a curve (dashed line) defined by the equation (fractionligand bound) = [N]/[N] + n/(Kw)], where [N] is the total lattice concentration.

I

9

10.’~

I

0

FIGURE 8: Same as Figure 7 except the concentration of total ligand is different in each figure: (A) LT = M; (B) L, = 5 X lo-’ M; (C) LT = 10” M.

region of ligand concentration (e.g., see Figure 2) where effectively no ligand binds. A consequence of the phenomenon described above is that it is potentially very easy to misinterpret the results of a titration in which lattice is beiiig added to ligand. An estimate of the site size, for example, based on the apparent end point, represents an underestimate of the true site size and, in addition, depends on the initial concentration of ligand used. If the experiment were a fluorescence titration in which the ligand fluorescence is quenched, the plateau value could correspond to an “apparent” value of the maximum quenching (Q,,,) that represents a (possibly major) underestimate of the true Q,,,. By using the apparent (incorrect) value of any quantitative analysis of an experimental titration curve would be meaningless and would yield incorrect binding parameters. Finally, if one were only drawing qualitative conclusions from such titration curves, their shapes might suggest that the system contains two types of binding sites with differing affinities, one with high affinity, corresponding to the steep region in Figures 7 and 8, and the other with low affinity, corresponding to the gradually increasing plateau region. In Figure 9 the effects of initial ligand concentration on the fractional saturation of the ligand, for a very highly cooperative system (w = lo7), are shown to illustrate more clearly the unique properties of these types of cooperative titration curves. Each line represents a calculated titration curve performed at a different initial ligand concentration. Because we have made w a very large value, the apparent plateaus seem very flat, but each curve actually does approach 100% saturation very slowly. On the basis of the previous discussion, the level of apparent saturation can be easily calculated as follows: since l / ( X w ) = 1 X IO-’ M, this concentration of ligand will remain unbound a t these lattice concentrations. Therefore, if the initial ligand concentration is 1 X then only .=(I X

e,

I

2

4

6

I 8

[PO y r A $ ( ~ M l

FIGURE 10: Titration of gene 32 protein with poly(rA) for different concentrationsof protein; percent fluorescencequenching of the protein is plotted against total poly(rA) concentration. These nonstoichiometric titrations were carried out in buffer C (Kowalczykowski et al., 1981b) containing 220 mM NaCl at 25 “C. Concentrations of gene 32 protein were 9.26 X lo-*, 1.85 X lo-’, and 6.48 X lo-’ M; the fitted curves were generated from eq 2 Nith parameters 12 = 7 . 5 , K = 9 X lo3 M-I, and w = 1.2 X lo3 by using a value of Qmax = 28.5% that represents 100%saturation of the ligand by poly(rA).

10-7)/(1 X = 90% of the total ligand will be bound to the lattice, which is the value of the apparent saturation plateau seen in Figure 9; a t 2.5 X initial ligand concentration, (1 X 10-7)/(2.5 X lo-’) = 60% will bind, etc. To illustrate that this type of behavior is actually observed in a “real” cooperative binding system, experimental titrations are presented in Figure 10 in which the quenching of the intrinsic protein fluorescence of gene 32 protein upon nucleic acid binding is monitored as poly(rA) is added to fixed concentrations of gene 32 protein. By changing the initial total concentration of gene 32 protein in the cuvette, curves with different apparent end points and different apparent values of are obtained (as expected from Figures 8 and 9). The true values of n and Qmaxwere determined by performing titrations under stoichiometric conditions a t low salt and were found to be 7.5 nucleotide residues and 28.5% (of the uncomplexed protein fluorescence), respectively. From comparison of the titration data with model curves, K was determined to be 9 X I O 3 M-l and w to be l .2 X IO3. These values of K and w are in good agreement with those obtained previously with ligand-to-lattice titration procedures (Jensen

e,,

et al., 1976; Kowalczykowski et al., 1981b; Newport et al., 1981). Thus, the proper analysis of this type of cooperative titration data requires that one know whether the observed plateau represents a true or an apparent end point; this is particularly important when indirect methods (such spectroscopic signals) are used to detect complex formation. The easiest way to determine whether a true end point has been reached in a lattice-to-ligand titration is to vary the initial ligand concentration (as in Figure 9) in order to see whether the apparent end-point plateau value is constant (when normalized for the difference in ligand concentrations) and whether the apparent site size is unchanged. If the value of n is invariant in each titration, then conditions are stoichiometric and the value of n is defined. If the apparent plateau values are invariant, but the experimentally determined values of n are not constant, then ligand is binding noncooperatively but binding conditions are not stoichiometric. Finally, if neither n nor the plateau value is invariant with initial ligand concentration, then binding is probably cooperative and results such as those of Figure 9 are expected. Once the value of n is known, the experimental titration curves can be analyzed to yield values of K and w by an approach similar to that described for ligand-to-lattice titrations. If the system is noncooperative, the value of K is varied until a good fit is obtained. If the system is highly cooperative (o> IO3), then one can take advantage of the fact that at the apparent “breakpoint” of a nonstoichiometric lattice-to-ligand titration the concentration of free ligand (not lattice) is approximately equal to l/(Ko). Then, as was done previously, model curves can be generated with a fixed value of Kw, but reciprocally changing values of K and w , until an adequate fit to the data is achieved. As with the ligand-to-lattice titration data, the experimental points may deviate from the calculated curves due to the fact that the experimental lattices are of finite size. However, in lattice-to-ligand titrations this deviation will manifest itself mostly at low lattice concentrations (the lefthand side of plots such as Figure 10). This is the region of the curve where the binding density is the greatest and where finite lattice effects are most pronounced (see below). Unfortunately, this is also the region of the titration curve that carries the most experimental “information”. Thus, latticeto-ligand titrations will tend to yield underestimated values of w .

Experiments at Very Low Binding Densities In addition to the titration methods that we have discussed, one can obtain values of K and w by any other method that allows the determination of the free ligand concentration. The approach to be described in this section takes advantage of the fact that at very low binding densities the amount of ligand bound to the lattice depends only on the value of K and is independent of w. An additional benefit of working at such low binding densities is that the “overlap” problem becomes insignificant, so that the amount bound will be independent of the site size, n. This is useful if stoichiometric binding conditions cannot be found for the system of interest. This effect is illustrated in Figure 11, in which we plot eq 2 with K held constant and calculate free ligand concentrations (L) as a function of u for different values of w . Note that at low binding densities (Le., v I all of the curves converge to a common value. In this region the free ligand concentration is independent of w , and from Figure 11 we see that this situation applies for binding densities (v) less than l/(lOw). The fact that the curves for constant values of K with variable values of w are superimposable at very low values of

1233

VOL. 2 5 , NO. 6 , 1986

BINDING PROTEINS TO NUCLEIC ACID LATTICES

- 4 -1

I

~

I

-10

1

-8

-6

-4

log

-2



FIGURE 11: Logarithm of the ligand concentration (Lr)vs. logarithm

of binding density ( u ) , for values of K = 1 M-’ and n = 7.

-

u allows one to determine a value of K alone, even for highly cooperative systems. By evaluating eq 2 in the limit as v 0, one obtains lim u/L = K (7) IC0

Thus, at very low values of u ( u < O.l/w), w and n drop out of eq 2, and L depends only on K. This approach can be exploited by any method in which low binding densities can be attained and free ligand concentration determined. For studies with T 4 gene 32 protein, a DNAcellulose column technique (deHaseth et al., 1977) was used, in which the concentration of bound DNA was approximately M (in nucleotide residues) and free protein concentrations were determined by intrinsic protein fluorescence, allowing values of u as low as to be attained [see Kowalczykowski et al. (198 1b) for details]. In addition, any other technique that enables one to work at high lattice-to-ligand ratios will be applicable [e.g., sedimentation (Draper & von Hippel, 1979; Yamamoto & Alberts, 1974) or ultrafiltration], coupled with any sensitive method for detecting the free ligand concentration, such as radioactive labeling, enzyme assay, or immunological detection; the lower limits of u and L attainable will depend only on the method used. Since eq 7 allows us to determine K independently of the effects of cooperativity, w cannot be determined directly from these low binding density experiments. However, the results can be coupled with an independent measurement of Kw to determine w as well. A method for determining Kw that differs from any described in this paper has been presented elsewhere (Kowalczykowski et al., 1981b; Newport et al., 1981) and referred to as the “salt back-titration” method (see references for details). Briefly, protein-nucleic acid complexes are formed and then titrated with NaCl to dissociate the complex. At salt concentrations at which half of the initially bound ligand is dissociated, KO = 1/L (if n